Alexis Roldan. Full-Stack Data Scientist & Sr. Data Engineer in Lancaster, CA.
Twenty years in biopharma manufacturing, thirteen of them in data roles.
I build manufacturing analytics and GenAI systems end to end — the ETL, the modeling, the app, the RAG layer, the deployment — on a global AI and data science team for Takeda's plasma manufacturing network, and I still own the Los Angeles site's two production platforms. The rule across all of them is the same: code computes the numbers, the model only explains them. Currently at Takeda Pharmaceutical.
alexis.m.roldan.ds@gmail.comWho I am and how I work.
I build production AI systems inside one of the most regulated manufacturing environments there is: human plasma fractionation. I started on the plant floor in 2005 as a manufacturing technician. Since August 2026 I work on a global AI and data science team for Takeda's plasma manufacturing network, and I still own the Los Angeles site's two production platforms — the ETL, the models, the apps, the RAG layer, the deployment and the executive deck.
What I'm known for is a specific architectural position: deterministic code computes every number, and the language model is only allowed to narrate it. In a GxP environment a model that invents a result isn't a bug, it's a compliance event. The constraint also cut token cost by about 70% as a side effect.
The work I'm proudest of usually involves removing something. Recently I measured the base rate of a feature everyone wanted, found the signal present in three records out of five, and deleted it. A flag that fires on 60% of your data isn't a flag.
R, SQL, Databricks, TypeScript. Twenty years in biopharma, thirteen in data roles.
English and Spanish (native), Japanese (conversational)
Plasma manufacturing, since 2005
Apps, pipelines and models, 13 years in data roles
AMIRA, manufacturing analytics
TOMO: statistics sent instead of rows
Race4Value — FTE Optimization Winner
2023, GMSGQ award across all plasma manufacturing sites
Star of the Quarter, Q3 FY22
2022, Los Angeles site
Employee of the Quarter
2017, Los Angeles site
Production platforms in regulated manufacturing, and the personal tools I build to keep learning.
20 projects, 10 with a full case study
R / Shiny
QC lab sample tracking with TV/kiosk mode and a forecast model — 3-phase SLA logic across 25 sample types. A proof of concept on dummy data, built so users can see the interface working before deciding whether to move forward.
Next.js / GenAI
Generates short-form reel concepts — hook → scene-by-scene → caption → CTA — streamed live, field-by-field from a song, vibe or lyric.
R / Shiny / NLP
R/Shiny text-analytics and search engine over TrackWise deviation events, fed daily from a Databricks pipeline: trending areas and functions by site, root-cause patterns with CAPA context, and keyword search across every event. 80% improvement in deviation search, closure and analytics, and the work behind the 2023 Race4Value award. Retired in 2026; the earlier generation of the question DevSight and Horizon later answered with retrieval.
R / Shiny / NLP
The TEASEr pattern applied to maintenance: JD Edwards work orders and their attachments wrangled in Databricks, then work-order analytics, equipment downtime from work-order cycle time, a preventive-maintenance calendar and keyword search over attachments, with text models retrained in an MLOps workflow. 75% less effort on work-order search, scheduling and trending. Retired in 2026.
R / Shiny / OCR
Interactive R/Shiny app extracting text from images with the Tesseract OCR engine — upload, select a region, extract for downstream analytics.
Python
Finds high-signal YouTube videos using view-to-subscriber ratio analysis — discovery without the algorithm.
R / Shiny / AWS
Personal dashboard surfacing latest videos from favorite channels — AWS-hosted data pipeline, deployed on Posit Connect.
Python
Python voice assistant with speech recognition, real-time financial data and weather integration.
R
Reusable analytics functions and custom visualization themes, installable via GitHub.
Twenty years in biopharma manufacturing, from the plant floor to a global AI and data science team.
Takeda – Global
Global AI / Data Science team (GMS) building GenAI and analytics solutions for Takeda's plasma manufacturing sites; still own AMIRA and BioTrack.
Takeda Pharmaceutical
Supported the Los Angeles Digital Roadmap, AI/ML development and advanced analytics to drive data-driven decisions.
Takeda Pharmaceutical
Drove the AGILE 4.0 Digital Workstream and cutting-edge analytics to accelerate Takeda's digital capabilities across GMS sites.
What I would be interviewed on, not everything I have touched.
Build Data Pipelines with Lakeflow Spark Declarative Pipelines — Databricks Academy
Data Warehousing with Databricks — Databricks Academy
Building Single-Agent Applications on Databricks — Databricks Academy
Oxford Machine Learning Summer School — AI for Global Goals
ODSC AI Bootcamp — Open Data Science Conference
Data Science Professional — HarvardX (Harvard Online)
Statistical Learning, Inference & Modeling — HarvardX / Stanford Online
B.S., Computer Science — Cal State Northridge
Ingeniero Informático, Informatics — Pontificia Universidad Católica del Perú
Hiring for a staff or lead data role, or need help with a regulated analytics or GenAI system? Email is fastest.
I'm open to conversations about data engineering, applied ML and GenAI in regulated and manufacturing environments. Write directly, or use the form and I'll reply from my inbox.
Lancaster, CA, Pacific time
